Measurement conditions
The figures shown come from real use. Here are the conditions under which they were recorded, and what they do not say.
Conditions of the reading
- 01Desktop machine powered on for several days, with other applications open.
- 02Both applications open on their instances list, at the same moment.
- 03The same gesture on both sides: open an instance's page — the same one — then go back to the list. The reading is taken at that precise moment.
- 04Sum of the memory of every process in the CurseForge group, as reported by the Windows task manager.
- 05No prior restart and no closing of other applications before measuring.
Recorded values
| CurseForge | Numek Launcher | |
|---|---|---|
| Processes | 7 to 15 | 1 |
| At rest | 400 to 600 MB | 10 MB |
| In use | 1,933 MB | 10 to 45 MB |
Supporting evidence
Both screenshots describe the same moment. They were taken one after the other, on the same machine, after opening the page of the same instance and returning to the list in each application.
8 processes · 1,933.2 MB · 65.5 % CPU
1 processes · 10.2 MB · 0.4 % CPU
Windows task manager. The CurseForge group is collapsed: the memory shown is the sum of its processes. The processor percentages are single-moment readings and are not averages. Numek Launcher was showing more instances than CurseForge at that moment, since it also displays CurseForge's own beside its: the gap counts against it.
Why not an isolated protocol
A benchmark on a freshly restarted machine would give more stable values, but less representative ones: nobody uses their computer in those conditions. The published values describe what a user observes day to day. In exchange they vary, which is why ranges are published rather than single values.
Why the values vary
Windows only reclaims a process's memory under physical memory pressure. An application that has navigated a lot therefore holds on to the peak it reached for a long time, and drops back only when the system needs that memory elsewhere. That is why the same application can be measured at 600 MB and then at over 1.9 GB on the same machine, depending on what happened in between. Numek Launcher does not rely on that pressure: a few seconds after opening, it hands back to the system the pages its startup called for and no longer uses. They only return if it needs them again, which explains both the low figure at rest and the range rather than a single value.
The two non-numeric rows
The advertising row can be verified by opening each application. The catalogues row can be verified by comparing the sources each one queries: Numek Launcher queries the public CurseForge and Modrinth programming interfaces.
What the comparison does not say
CurseForge offers functions Numek Launcher does not, and its advertising funds payouts to mod authors. The items above concern resource usage and two verifiable characteristics, and nothing else.